pvary commented on a change in pull request #1495:
URL: https://github.com/apache/iceberg/pull/1495#discussion_r494088866
##########
File path: mr/src/main/java/org/apache/iceberg/mr/hive/HiveIcebergSerDe.java
##########
@@ -27,21 +27,32 @@
import org.apache.hadoop.hive.serde2.SerDeStats;
import org.apache.hadoop.hive.serde2.objectinspector.ObjectInspector;
import org.apache.hadoop.io.Writable;
-import org.apache.iceberg.Table;
+import org.apache.iceberg.Schema;
+import org.apache.iceberg.SchemaParser;
+import org.apache.iceberg.exceptions.NoSuchTableException;
import org.apache.iceberg.mr.Catalogs;
+import org.apache.iceberg.mr.InputFormatConfig;
import org.apache.iceberg.mr.hive.serde.objectinspector.IcebergObjectInspector;
import org.apache.iceberg.mr.mapred.Container;
public class HiveIcebergSerDe extends AbstractSerDe {
-
+ private Schema schema;
private ObjectInspector inspector;
@Override
public void initialize(@Nullable Configuration configuration, Properties
serDeProperties) throws SerDeException {
- Table table = Catalogs.loadTable(configuration, serDeProperties);
-
try {
- this.inspector = IcebergObjectInspector.create(table.schema());
+ String schemaString = (String)
serDeProperties.get(InputFormatConfig.TABLE_SCHEMA);
+ if (schemaString != null) {
+ schema = SchemaParser.fromJson(schemaString);
+ } else {
+ try {
+ schema = Catalogs.loadTable(configuration, serDeProperties).schema();
+ } catch (NoSuchTableException nte) {
+ throw new SerDeException("Please provide an existing table or a
valid schema", nte);
+ }
+ }
+ inspector = IcebergObjectInspector.create(schema);
Review comment:
Added a comment:
HiveIcebergSerDe.initialize is called multiple places in Hive code:
- When we are trying to create a table - HiveDDL data is stored at the
serDeProperties, but no Iceberg table is created yet.
- When we are compiling the Hive query on HiveServer2 side - We only have
table information (location/name), and we have to read the schema using the
table data. This is called multiple times so there is room for optimizing here.
- When we are executing the Hive query in the execution engine - We do not
want to load the table data on every executor, but serDeProperties are
populated by HiveIcebergStorageHandler.configureInputJobProperties() and the
resulting properties are serialized and distributed to the executors
----------------------------------------------------------------
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
